Future rainfall prediction method, device and system and storage medium

Abstract

The invention discloses a future rainfall prediction method, device and system and a storage medium, and belongs to the technical field of hydrometeorological forecasting, and the method comprises the steps: constructing a multi-site Neyman-Scott rectangular pulse random rainfall model based on the actual measurement data of ground rainfall, optimizing the parameters, and obtaining the optimized multi-site Neyman-Scott rectangular pulse random rainfall model; obtaining a disturbance change factor based on historical rainfall test data and future rainfall simulation data of the optimal GCM, and obtaining a statistical magnitude corresponding to future rainfall prediction data based on the disturbance change factor; and inputting the statistical magnitude corresponding to the future rainfall prediction data into the optimized multi-site Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data. According to the method, extreme features of rainfall and spatial correlation features of multi-site rainfall can be accurately simulated, and the accuracy and reliability of future rainfall prediction are improved.

Description

Technical Field

[0001] The invention relates to a future rainfall prediction method, device, system and storage medium, belonging to the technical field of hydrological and meteorological forecasting. Background Art

[0002] The Global Climate Model (GCM) is the main tool for predicting future rainfall. However, the spatial resolution of GCM output data is low and there are biases. Using downscaling methods to convert GCM large-scale coarse-resolution climate variables to regional ones is the key to improving the credibility of future rainfall predictions.

[0003] Downscaling methods are divided into two categories: dynamical downscaling and statistical downscaling. Dynamical downscaling is to embed a higher-resolution regional climate model (RCM) in the GCM. The GCM provides initial and boundary conditions for the RCM, and numerically integrates the high-resolution regional climate model to obtain high-resolution climate information. However, dynamical downscaling strongly relies on the boundary forcing of the GCM, inherits the bias and defects of the GCM, and its prediction results need to be further statistically downscaled. The computational cost is high, and the number of application areas and scenarios is limited. Statistical downscaling is to construct the statistical relationship between the large-scale climate elements of the GCM and the regional measured climate elements. Compared with dynamical downscaling, statistical downscaling has low computational cost and high efficiency, and is convenient for outputting point-scale climate variables.

[0004] Statistical downscaling can be divided into three categories: regression models, circulation typing techniques, and random weather generators. Regression models, or transfer functions, are direct quantitative functional relationships between regional climate variables (such as temperature and precipitation) and a set of large-scale climate variables. Circulation typing techniques classify large-scale atmospheric circulation types related to regional climate change, link the occurrence of specific weather types to regional climate variables, and use GCM to predict changes in the frequency of weather types to predict climate change. Random weather generators (Weather Generators, WG) obtain statistical characteristic values ​​based on the measured values ​​of climate variables, and then use random number generators to construct random processes of climate elements based on the statistical characteristic values. In future climate predictions, changes in climate variability are more important than changes in the mean, so WG is widely used to generate future climate change scenarios, but WG has weak simulation capabilities for extreme values.

[0005] Therefore, there is an urgent need for a downscaling model that can simulate extreme rainfall more accurately and improve the accuracy and reliability of future rainfall forecasts. Summary of the invention

[0006] The purpose of the present invention is to provide a method, device, system and storage medium for predicting future rainfall, which can accurately simulate the extreme characteristics of rainfall and the spatial correlation characteristics of rainfall at multiple sites, and improve the accuracy and reliability of future rainfall prediction.

[0007] In order to achieve the above object, the present invention provides the following technical solutions: In a first aspect, the present invention provides a method for predicting future rainfall, comprising: Obtain ground rainfall measured data, historical rainfall test data of the best GCM, and future rainfall simulation data; Based on the measured rainfall data, a multi-site Neyman-Scott rectangular pulse random rainfall model is constructed, and the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model. Based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, the disturbance change factor is obtained, and based on the disturbance change factor, the statistics corresponding to the future rainfall prediction data are obtained; The statistics corresponding to the future rainfall prediction data are input into the optimized multi-station Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data; Among them, the optimal GCM is the GCM with the highest simulation ability screened out after evaluating the simulation ability of several GCMs.

[0008] Combined with the first aspect, we further evaluated the simulation capabilities of several GCMs and screened out the GCMs with the highest simulation capabilities, including: Calculate the simulation capability score of GCM under each evaluation index; Based on the simulation ability scores of GCM under each evaluation index, the total simulation ability score of GCM is calculated; Based on the total simulation ability score of each GCM, the GCM with the highest simulation ability was selected; Among them, the evaluation indicators include relative error evaluation indicators and non-relative error evaluation indicators.

[0009] In combination with the first aspect, further, the relative error evaluation index includes: The relative error of the mean of GCM historical rainfall test data to the mean of ground rainfall measured data; The relative error of the coefficient of variation of GCM historical rainfall test data relative to the coefficient of variation of ground rainfall measured data; The normalized root mean square error of the GCM historical rainfall test data relative to the ground rainfall measured data; The relative error between the change trend of GCM historical rainfall test data in the Mann-Kendall trend analysis and the change trend of ground rainfall measured data in the Mann-Kendall trend analysis; The relative error between the trend value of the GCM historical rainfall test data in the Mann-Kendall trend analysis and the trend value of the ground rainfall measured data in the Mann-Kendall trend analysis; Non-relative error evaluation indicators include: The temporal correlation coefficient between the historical rainfall test data of GCM and the measured rainfall data on the ground; Spatial correlation coefficient of GCM historical rainfall test data with ground rainfall measured data.

[0010] Combined with the first aspect, further, the calculation formula of the simulation ability score of GCM under each evaluation index is: ; in, Indicates that GCM is The simulation ability score under the evaluation index is Indicates the GCM corresponding The value of the evaluation index, , Indicates the first The maximum and minimum values ​​of the evaluation indicators; The calculation formula for the total simulation ability score of GCM is: ; in, It indicates the total simulation ability score of GCM. The higher the total simulation ability score, the higher the simulation ability. express The corresponding weight, Represents the total number of evaluation indicators.

[0011] Combined with the first aspect, further, based on the measured rainfall data on the ground, a multi-station Neyman-Scott rectangular pulse random rainfall model is constructed, including: Based on the statistics corresponding to the measured rainfall data on the ground and the fitting weights corresponding to the statistics, the parameters of the multi-station Neyman-Scott rectangular pulse random rainfall model are fitted. Based on the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model, a multi-site Neyman-Scott rectangular pulse random rainfall model is generated; Among them, the statistics include rainfall mean, variance, cross-correlation coefficient of each station, dry day probability and skewness coefficient; Generate a multi-site Neyman-Scott rectangular pulse random rainfall model including: The starting points of rainfall events are randomly generated according to a Poisson process; A random number of rainfall units are generated within each rainfall event. The intervals between the starting points of rainfall units after adjacent rainfall events are independent of each other and follow an exponential distribution. The rainfall units within the same rainfall event follow a Poisson distribution in space, an exponential distribution in radius, and independent duration and rainfall intensity and follow an exponential distribution. The parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model include the characterization parameters of the average interval time between the starting points of adjacent rainfall events, the characterization parameters of the average number of rainfall units within each rainfall event, the characterization parameters of the average interval time between the starting points of rainfall units after adjacent rainfall events, the characterization parameters of the spatial distribution average density of rainfall units within the same rainfall event, the characterization parameters of the average radius of rainfall units within the same rainfall event, the characterization parameters of the average duration of rainfall units within the same rainfall event, the characterization parameters of the average rainfall intensity of rainfall units within the same rainfall event and the scale factor; The value of the scale factor is the quantile of the average rainfall at each station.

[0012] In combination with the first aspect, further, the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model, including: Based on the objective function, the parameters of the multi-station Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-station Neyman-Scott rectangular pulse random rainfall model; Among them, the objective function is: ; in, Indicates the first Statistics, express The corresponding annual average value is express The corresponding fitting weight is represents the set of parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model, The multi-station Neyman-Scott rectangular pulse random rainfall model corresponds to The output ground rainfall simulation data corresponds to the Statistics, Indicates the total number of statistics.

[0013] Combined with the first aspect, further, the calculation formula of the disturbance change factor is: ; in, Indicates The disturbance change factor corresponding to the statistic is , The historical rainfall test data and future rainfall simulation data of the optimal GCM correspond to the Statistics; The calculation formula of the statistics corresponding to the future rainfall forecast data is: ; in, Indicates the future rainfall forecast data corresponding to the Statistics, Indicates the first A statistic.

[0014] In a second aspect, the present invention provides a future rainfall prediction device, comprising: The data acquisition module is used to obtain the measured rainfall data on the ground, the historical rainfall test data of the optimal GCM, and the future rainfall simulation data; A model building module is used to build a multi-site Neyman-Scott rectangular pulse random rainfall model based on ground rainfall measured data, and optimize the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model to obtain an optimized multi-site Neyman-Scott rectangular pulse random rainfall model; The downscaling module is used to obtain the disturbance change factor based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, and obtain the statistics corresponding to the future rainfall prediction data based on the disturbance change factor; A prediction module is used to input the statistics corresponding to the future rainfall prediction data into the optimized multi-site Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data; Among them, the optimal GCM is the GCM with the highest simulation ability screened out after evaluating the simulation ability of several GCMs.

[0015] In a third aspect, the present invention provides a computer system, comprising: Storage medium: used to store computer programs; Processor: used to execute the computer program to implement the steps of the future rainfall prediction method described in any one of the first aspects.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the future rainfall prediction method described in any one of the first aspects.

[0017] In a fifth aspect, the present invention provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the future rainfall prediction method described in any one of the first aspects.

[0018] Compared with the prior art, the present invention has the following beneficial effects: The future rainfall prediction method provided by the present invention adopts an optimized multi-site Neyman-Scott rectangular pulse random rainfall model, combined with a disturbance change factor based on the optimal GCM, to predict future rainfall, and uses a random point process to simulate rainfall, taking uncertainty into account, and can better simulate the extreme characteristics of rainfall. Introducing spatial distribution parameters and jointly predicting each site in the basin can more accurately simulate the spatial correlation characteristics of multi-site rainfall, improve the prediction accuracy, and can predict rainfall for sites with insufficient data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a flow chart of a future rainfall prediction method provided by an embodiment of the present invention; Figure 2 It is a schematic diagram of a multi-site Neyman-Scott rectangular pulse random rainfall model provided by an embodiment of the present invention; Figure 3 It is a relative error spatial distribution diagram of the original EC-Earth3 model in the validation period provided by the embodiment of the present invention and the EC-Earth3 model downscaled by the Delta method, NSRP, and STNSRP relative to the measured data; Figure 4 It is a comparison chart of the monthly statistical characteristics of the measured rainfall in the whole basin and Liyang station during the validation period and the simulated rainfall after downscaling using the Delta method, NSRP, and STNSRP; Figure 5 is a box plot of monthly precipitation changes in the three future periods relative to the historical period provided by an embodiment of the present invention; Figure 6 It is a schematic diagram of rainfall changes predicted by various modes in the next three periods provided by an embodiment of the present invention; Figure 7 is a spatial distribution diagram of the change of the average annual rainfall in the next three periods relative to the reference period predicted by each model provided in the embodiment of the present invention; Figure 8 This is a schematic diagram of the annual changes in the extreme rainfall index in the Taihu Lake Basin from 2021 to 2100 relative to the base period provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0020] The technical solution of the present application is further described in detail below in conjunction with specific implementation methods.

[0021] The embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present application, and should not be construed as limitations on the present application. In the absence of conflict, the embodiments of the present application and the technical features in the embodiments may be combined with each other.

[0022] Embodiment 1: This embodiment provides a method for predicting future rainfall. Figure 1 is a flow chart of the future rainfall prediction method provided by this embodiment. The flow chart only shows the logical sequence of the method of this embodiment. Different methods can be used without conflict. Figure 1 The steps shown or described are accomplished in the order shown.

[0023] The future rainfall prediction method provided in this embodiment can be applied to a terminal and can be executed by a future rainfall prediction device. The device can be implemented by software and / or hardware, and the device can be integrated in a terminal, for example: any tablet computer or computer device with communication function.

[0024] like Figure 1 As shown, the future rainfall prediction method provided in this embodiment specifically includes the following steps: Obtain ground rainfall measured data, historical rainfall test data of the best GCM, and future rainfall simulation data; Based on the measured rainfall data, a multi-site Neyman-Scott rectangular pulse random rainfall model is constructed, and the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model. Based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, the disturbance change factor is obtained, and based on the disturbance change factor, the statistics corresponding to the future rainfall prediction data are obtained; The statistics corresponding to the future rainfall prediction data are input into the optimized multi-station Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data; Among them, the optimal GCM is the GCM with the highest simulation ability screened out after evaluating the simulation ability of several GCMs.

[0025] Embodiment 2: This embodiment provides a future rainfall prediction method. Based on Embodiment 1, the future rainfall prediction method provided by this embodiment specifically includes the following steps: Step 1: Obtain ground rainfall measured data, historical rainfall test data of the best GCM, and future rainfall simulation data; In this embodiment, the optimal GCM is a GCM with the highest simulation capability that is screened out after evaluating the simulation capabilities of several GCMs.

[0026] The simulation capability evaluation of several GCMs and the selection of the GCM with the highest simulation capability include the following steps: Step 1: Calculate the simulation capability score of GCM under each evaluation index; In this embodiment, the evaluation index includes a relative error evaluation index and a non-relative error evaluation index.

[0027] The relative error evaluation indicators specifically include: (1) The relative error of the mean of the GCM historical rainfall test data to the mean of the ground rainfall measured data; In this embodiment, the calculation formula of the mean is: ; in, represents the mean of the historical rainfall test data of GCM, Indicates the historical rainfall test data of GCM The value corresponding to the time series; represents the mean of the measured rainfall data on the ground. Indicates the measured rainfall data on the ground The value corresponding to the time series, It represents the total number of time series of historical rainfall experiment data of GCM or ground rainfall measurement data.

[0028] Specifically, in the calculation formula of the mean, The time series can represent the Day Month or Year.

[0029] (2) The relative error of the coefficient of variation of the GCM historical rainfall test data relative to the coefficient of variation of the ground rainfall measured data; In this embodiment, the calculation formula of the coefficient of variation is: ; in, represents the coefficient of variation of GCM historical rainfall data, It represents the coefficient of variation of the measured rainfall data.

[0030] Specifically, in the calculation formula of the coefficient of variation, The time series can represent the Day Month or Year.

[0031] (3) The normalized root mean square error of the GCM historical rainfall test data relative to the ground rainfall measured data; In this embodiment, the calculation formula of the standardized root mean square error of the GCM historical rainfall test data relative to the ground rainfall measured data is: ; in, It represents the normalized root mean square error of the GCM historical rainfall test data relative to the ground rainfall measured data.

[0032] Specifically, in the calculation formula of the standardized root mean square error of the GCM historical rainfall test data relative to the ground rainfall measured data, The time series can represent the Day Month or Year.

[0033] (4) The relative error between the change trend of the GCM historical rainfall test data in the Mann-Kendall trend analysis and the change trend of the ground rainfall measured data in the Mann-Kendall trend analysis; In this embodiment, the calculation formula of the change trend is: ; in, Indicates the trend of change, The table seems to be changing in an increasing trend. Indicates that the trend of change is unchanged. Indicates that the trend of change is decreasing. represents the intermediate variable, when When expressing the change trend of GCM historical rainfall test data in Mann-Kendall trend analysis, ,in, Indicates the historical rainfall test data of GCM The value corresponding to the time series is When the measured rainfall data on the ground are expressed in the Mann-Kendall trend analysis, ,in, Indicates the measured rainfall data on the ground The value corresponding to the time series.

[0034] Specifically, in the intermediate variable In the expression of The time series can represent the Day Month or Year, No. The time series can represent the Day Month or Year.

[0035] (5) The relative error between the trend value of the GCM historical rainfall test data in the Mann-Kendall trend analysis and the trend value of the ground rainfall measured data in the Mann-Kendall trend analysis.

[0036] In this embodiment, the calculation formula of the trend value is: ; in, It represents the trend value corresponding to the historical rainfall test data of GCM in the Mann-Kendall trend analysis. It indicates the trend value corresponding to the measured ground rainfall data in the Mann-Kendall trend analysis.

[0037] Specifically, in the calculation formula of the trend value, The time series can represent the Day Month or Year, No. The time series can represent the Day Month or Year.

[0038] The non-relative error evaluation indicators specifically include: (1) The temporal correlation coefficient between the GCM historical rainfall test data and the ground rainfall measured data; In this embodiment, the calculation formula of the time correlation coefficient of the historical rainfall test data of GCM with respect to the ground rainfall measured data is: ; in, It represents the temporal correlation coefficient between the historical rainfall test data of GCM and the measured rainfall data on the ground.

[0039] Specifically, in the calculation formula of the time correlation coefficient, The time series can represent the Day Month or Year.

[0040] (2) Spatial correlation coefficient between GCM historical rainfall test data and ground rainfall measured data.

[0041] In this embodiment, the calculation formula of the spatial correlation coefficient of the GCM historical rainfall test data with respect to the ground rainfall measured data is: ; in, It represents the spatial correlation coefficient between the historical rainfall test data of GCM and the measured rainfall data on the ground. Indicates the historical rainfall test data of GCM The value corresponding to each site, represents the mean of GCM's historical rainfall test data for each station, Indicates the measured rainfall data on the ground The value corresponding to each site, represents the mean of the measured rainfall data at each station. The total number of stations that represent the historical rainfall experiment data of GCM or the ground rainfall measurement data.

[0042] Specifically, , , , They all correspond to the same time series, such as the same year.

[0043] In this embodiment, the calculation formula for the simulation ability score of GCM under each evaluation index is: ; in, Indicates that GCM is The simulation ability score under the evaluation index is Indicates the GCM corresponding The value of the evaluation index, , Indicates the first The maximum and minimum values ​​of the evaluation indicators.

[0044] Step 2: Based on the simulation ability scores of GCM under each evaluation index, calculate the total simulation ability score of GCM; In this embodiment, the calculation formula for the total score of the simulation capability of the GCM is: ; in, It indicates the total simulation ability score of GCM. The higher the total simulation ability score, the higher the simulation ability. express The corresponding weight, Represents the total number of evaluation indicators.

[0045] In this embodiment, the total number of evaluation indicators is 7, the first evaluation index is the relative error of the mean of GCM historical rainfall test data to the mean of ground rainfall measured data, the second evaluation index is the relative error of the coefficient of variation of GCM historical rainfall test data to the coefficient of variation of ground rainfall measured data, the third evaluation index is the standardized root mean square error of GCM historical rainfall test data to ground rainfall measured data, the fourth evaluation index is the relative error of the change trend of GCM historical rainfall test data in Mann-Kendall trend analysis to the change trend of ground rainfall measured data in Mann-Kendall trend analysis, the fifth evaluation index is the relative error of the trend value of GCM historical rainfall test data in Mann-Kendall trend analysis to the trend value of ground rainfall measured data in Mann-Kendall trend analysis, the sixth evaluation index is the time correlation coefficient of GCM historical rainfall test data to ground rainfall measured data, the seventh evaluation index is the spatial correlation coefficient of GCM historical rainfall test data to ground rainfall measured data, correspondingly, , , , , is 1 / 6, , is 1 / 12.

[0046] Step 3: Based on the total simulation ability scores of each GCM, select the GCM with the highest simulation ability.

[0047] In this embodiment, the higher the total simulation capability score of the GCM is, the higher the simulation capability of the GCM is; the GCM with the highest total simulation capability score is the GCM with the highest simulation capability.

[0048] Compared with other evaluation methods such as Taylor diagram, the GCM simulation capability evaluation method provided in this embodiment takes into account more dimensional statistics, can more comprehensively reflect the degree of fit between simulated variables and measured variables, and the evaluation results are more intuitive and objective, which is more conducive to comprehensive evaluation.

[0049] Step 2: Based on the measured rainfall data on the ground, a multi-site Neyman-Scott rectangular pulse random rainfall model is constructed, and the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain an optimized multi-site Neyman-Scott rectangular pulse random rainfall model; The Neyman-Scott rectangular pulse (NSRP) model is a model that randomly generates rainfall based on a point process. It simulates the occurrence of rainfall events, the intensity of rainfall units, and the distribution of rainfall time. It is based on the clustering characteristics of rainfall units and uses relatively few parameters to reflect the rainfall process and underlying physical phenomena. It considers the rainfall extreme value problem based on the aggregation time of different rainfalls and can accurately simulate the temporal variation of rainfall.

[0050] This embodiment introduces spatial distribution parameters into the NSRP model, fully considers the spatial correlation characteristics of rainfall, performs joint prediction on each station in the basin, and proposes a multi-station Neyman-Scott rectangular pulse random rainfall (STNSRP) model.

[0051] In this embodiment, based on the measured rainfall data on the ground, constructing a multi-site Neyman-Scott rectangular pulse random rainfall model specifically includes the following steps: Step 1: Based on the statistics corresponding to the measured rainfall data and the fitting weights corresponding to the statistics, the parameters of the multi-station Neyman-Scott rectangular pulse random rainfall model are fitted; In this embodiment, the statistics include rainfall mean, variance, correlation coefficient, dry day probability and skewness coefficient. The statistics and the fitting weights corresponding to the statistics are shown in Table 1.

[0052] Table 1 Statistics and their corresponding fitting weights

[0053] As can be seen from Table 1, for a single site, the correlation coefficient is the autocorrelation coefficient of the single site; for multiple sites, the correlation coefficient is the mutual correlation coefficient of each site.

[0054] Specifically, the statistics corresponding to the measured rainfall data include the mean, variance, the correlation coefficient of each station, the dry day probability and the skewness coefficient corresponding to the measured rainfall data.

[0055] In this embodiment, the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model include a characterization parameter of the average interval time between the starting points of adjacent rainfall events, a characterization parameter of the average number of rainfall units within each rainfall event, a characterization parameter of the average interval time between the starting points of rainfall units after adjacent rainfall events, a characterization parameter of the spatial distribution average density of rainfall units within the same rainfall event, a characterization parameter of the average radius of rainfall units within the same rainfall event, a characterization parameter of the average duration of rainfall units within the same rainfall event, a characterization parameter of the average rainfall intensity of rainfall units within the same rainfall event, and a scale factor. The parameter list of the multi-site Neyman-Scott rectangular pulse random rainfall model is shown in Table 2.

[0056] Table 2 Parameters of the multi-station Neyman-Scott rectangular pulse random rainfall model

[0057] In this embodiment, in order to consider the influence of the terrain factors of each site, a scale factor is introduced , scale factor The value of is the quantile of the average rainfall at each station.

[0058] Step 2: Generate a multi-site Neyman-Scott rectangular pulse random rainfall model based on the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model.

[0059] The multi-site Neyman-Scott rectangular pulse random rainfall model simulates rainfall events through three random processes: the first is the process of determining the starting position of the rainfall event, the second is the process of determining the number of rainfall units within the rainfall event, and the third is the process of determining the location of the rainfall unit. After determining the starting position of the rainfall event, the number of rainfall units within the rainfall event, and the location of the rainfall unit, the rainfall unit has a random rainfall duration and rainfall intensity, which can reflect the temporal variation of rainfall intensity.

[0060] In this embodiment, generating a multi-site Neyman-Scott rectangular pulse random rainfall model specifically includes the following steps: Step ①: Randomly generate the starting point of the rainfall event according to the Poisson process; In this embodiment, Figure 2 As shown in (a), the average interval between the starting points of adjacent rainfall events is expressed as express. Figure 2 In the figure, red dots indicate rainfall events.

[0061] Step ②: Generate a random number of rainfall units within each rainfall event. The intervals between the starting points of rainfall units after adjacent rainfall events are independent of each other and follow an exponential distribution. The rainfall units within the same rainfall event follow a Poisson distribution in space, an exponential distribution in radius, and independent duration and rainfall intensity and follow an exponential distribution.

[0062] In this embodiment, Figure 2 As shown in (b), the average number of rainfall cells within each rainfall event is expressed as The average interval time between the starting points of rainfall units after adjacent rainfall events is expressed as The spatial average density of rainfall units within the same rainfall event is expressed as The average radius is expressed as The average duration is expressed as The average rainfall intensity is expressed as express. Figure 2 In the figure, red five-pointed stars represent rainfall cells.

[0063] In this embodiment, Figure 2 As shown in (c), the rainfall intensity of each rainfall event is equal to the sum of the rainfall intensities of all rainfall units within it during their life cycle.

[0064] In this embodiment, the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model, which specifically includes: based on the objective function, the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model.

[0065] In this embodiment, the objective function is: ; in, Indicates the first Statistics, express The corresponding annual average value is express The corresponding fitting weight is represents the set of parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model, The multi-station Neyman-Scott rectangular pulse random rainfall model corresponds to The output ground rainfall simulation data corresponds to the Statistics, Indicates the total number of statistics.

[0066] Specifically, the statistics corresponding to the measured ground rainfall data include the mean, variance, mutual correlation coefficient of each station, dry day probability and skewness coefficient corresponding to the measured ground rainfall data; the statistics corresponding to the simulated ground rainfall data include the mean, variance, mutual correlation coefficient of each station, dry day probability and skewness coefficient corresponding to the simulated ground rainfall data.

[0067] Step 3: Based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, obtain the disturbance change factor, and based on the disturbance change factor, obtain the statistics corresponding to the future rainfall prediction data; The role of the historical rainfall test data and future rainfall simulation data of the optimal GCM is to provide the STNSRP model with disturbance information of climate change. Assuming that the future rainfall changes of each station are proportional to the changes between the historical rainfall test data and future rainfall simulation data of the GCM, the disturbance change factors of each statistic of each station in the future period relative to the historical period are calculated based on the historical rainfall test data and future rainfall simulation data of the GCM. Since the GCM has a poor description of the statistic of the mutual correlation coefficient, the change factor of the correlation coefficient is not calculated, and the historical data is directly used.

[0068] In this embodiment, the calculation formula of the disturbance change factor is: ; in, Indicates The disturbance change factor corresponding to the statistic is , The historical rainfall test data and future rainfall simulation data of the optimal GCM correspond to the A statistic.

[0069] Specifically, the statistics corresponding to the historical rainfall test data of the optimal GCM include the mean, variance, mutual correlation coefficient of each station, dry day probability and skewness coefficient corresponding to the historical rainfall test data of the optimal GCM; the statistics corresponding to the future rainfall simulation data of the optimal GCM include the mean, variance, mutual correlation coefficient of each station, dry day probability and skewness coefficient corresponding to the future rainfall simulation data of the optimal GCM.

[0070] In this embodiment, the calculation formula of the statistics corresponding to the future rainfall prediction data is: ; in, Indicates the future rainfall forecast data corresponding to the Statistics, Indicates the first A statistic.

[0071] Specifically, the statistics corresponding to the future rainfall prediction data include the mean, variance, mutual correlation coefficient of each station, dry day probability and skewness coefficient corresponding to the future rainfall prediction data.

[0072] Step 4: Input the statistics corresponding to the future rainfall prediction data into the optimized multi-site Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data.

[0073] In this embodiment, the statistics corresponding to the future rainfall prediction data are input into the optimized multi-site Neyman-Scott rectangular pulse random rainfall model, and the optimized multi-site Neyman-Scott rectangular pulse random rainfall model outputs the future rainfall prediction data based on the statistics corresponding to the future rainfall prediction data.

[0074] Embodiment 3: This embodiment provides a method for predicting future rainfall. Based on Embodiment 2, obtaining ground rainfall measured data and historical rainfall test data of the best GCM and future rainfall simulation data specifically includes the following steps: Step A: Obtain ground rainfall measured data and several GCM historical rainfall test data and future rainfall simulation data; In this embodiment, the measured ground rainfall data are collected from the ground rain gauges in the study area, and the historical rainfall test data and future rainfall simulation data of a preset number (preset number ≥ 2) GCMs are collected from the Sixth International Coupled Model Intercomparison Project Phase 6 (CMIP6).

[0075] Specifically, the measured rainfall data on the ground, the historical rainfall test data of each GCM, and the future rainfall simulation data are all daily data.

[0076] Step B: Based on the measured rainfall data on the ground, preprocess the historical rainfall test data and future rainfall simulation data of each GCM; In this embodiment, based on the measured rainfall data on the ground, preprocessing the historical rainfall test data and future rainfall simulation data of each GCM specifically includes the following steps: Step a: According to the period of the ground rainfall measured data, the period of the historical rainfall test data of each GCM is intercepted to make the period of the historical rainfall test data of each GCM consistent with the period of the ground rainfall measured data; Step b: According to the station location of the ground rainfall measurement data, the historical rainfall test data and future rainfall simulation data of each GCM are interpolated to the station of the ground rainfall measurement data using the inverse distance interpolation method.

[0077] Step C: Based on the measured rainfall data and the pre-processed historical rainfall test data of each GCM, the simulation capability of each GCM is evaluated, and the GCM with the highest simulation capability is selected as the optimal GCM; In this embodiment, based on the measured rainfall data on the ground and the pre-processed historical rainfall test data of each GCM, the simulation capability of each GCM is evaluated, and the GCM with the highest simulation capability is selected as the optimal GCM, which specifically includes the following steps: Step a: Based on the measured rainfall data on the ground and the pre-processed historical rainfall test data of each GCM, the values ​​of each evaluation index corresponding to each GCM are calculated; In this embodiment, based on the ground rainfall measured data and the pre-processed historical rainfall test data of each GCM, the calculation formulas for the mean, coefficient of variation, change trend, and trend value provided in Example 2 are used to calculate the mean, coefficient of variation, change trend corresponding to the Mann-Kendall trend analysis, and trend value corresponding to the Mann-Kendall trend analysis of the historical rainfall test data of each GCM, as well as the mean, coefficient of variation, change trend corresponding to the Mann-Kendall trend analysis, and trend value corresponding to the Mann-Kendall trend analysis of the ground rainfall measured data, and then calculate the historical rainfall test data of each GCM. The relative error value of the mean of the historical rainfall test data relative to the mean of the ground rainfall measured data, the relative error value of the coefficient of variation of the historical rainfall test data of each GCM relative to the coefficient of variation of the ground rainfall measured data, the relative error value of the corresponding changing trend of the historical rainfall test data of each GCM in the Mann-Kendall trend analysis relative to the corresponding changing trend of the ground rainfall measured data in the Mann-Kendall trend analysis, and the relative error value of the corresponding trend value of the historical rainfall test data of each GCM in the Mann-Kendall trend analysis relative to the corresponding trend value of the ground rainfall measured data in the Mann-Kendall trend analysis.

[0078] Based on the measured ground rainfall data and the preprocessed historical rainfall test data of each GCM, the standardized root mean square error of the historical rainfall test data of GCM relative to the measured ground rainfall data, the time correlation coefficient of the historical rainfall test data of GCM relative to the measured ground rainfall data, and the calculation formula for the spatial correlation coefficient of the historical rainfall test data of GCM relative to the measured ground rainfall data provided in Example 2 are used to calculate the value of the standardized root mean square error of the historical rainfall test data of each GCM relative to the measured ground rainfall data, the value of the time correlation coefficient of the historical rainfall test data of each GCM relative to the measured ground rainfall data, and the value of the spatial correlation coefficient of the historical rainfall test data of each GCM relative to the measured ground rainfall data.

[0079] Step b: Based on the values ​​of each evaluation index corresponding to each GCM, the simulation capability score of each GCM under each evaluation index is calculated; In this embodiment, based on the values ​​of each evaluation index corresponding to each GCM, the maximum and minimum values ​​of the values ​​of each evaluation index corresponding to each GCM are calculated, and then the simulation capability score of each GCM under each evaluation index is calculated using the calculation formula for the simulation capability score of the GCM under each evaluation index provided in Example 2.

[0080] Step c: Based on the simulation ability scores of each GCM under each evaluation index, the total simulation ability score of each GCM is calculated; In this embodiment, based on the simulation capability score of each GCM under each evaluation index, the calculation formula for the total simulation capability score of the GCM provided in Example 2 is used to calculate the total simulation capability score of each GCM.

[0081] Step d: Based on the total simulation ability scores of each GCM, the GCM with the highest simulation ability is selected as the optimal GCM.

[0082] Step D: Obtain historical rainfall test data and future rainfall simulation data of the optimal GCM.

[0083] In this embodiment, the historical rainfall test data and future rainfall simulation data of the optimal GCM are both pre-processed historical rainfall test data and future rainfall simulation data.

[0084] The future rainfall prediction method provided in this embodiment pre-processes the historical rainfall test data and future rainfall simulation data of each GCM before evaluating the simulation capability of each GCM, so that the time period of the historical rainfall test data of each GCM is consistent with the ground rainfall measured data, and interpolates the historical rainfall test data and future rainfall simulation data of each GCM to the site of the ground rainfall measured data, so as to compare the ground rainfall measured data with the historical rainfall test data of each GCM.

[0085] Embodiment 4: This embodiment provides a method for predicting future rainfall. Based on Embodiment 2, the method for predicting future rainfall provided by this embodiment can be applied to predicting future rainfall in the Taihu Lake Basin.

[0086] The future rainfall prediction method provided in this embodiment specifically includes the following steps: (1) Multi-source rainfall data collection and preprocessing.

[0087] Daily rainfall data from 20 surface rainfall stations in the Taihu Lake Basin from 1954 to 2019 were collected. The site information of the surface rainfall stations is shown in Table 3.

[0088] Table 3 Site information of surface rainfall stations

[0089] The historical experimental data of 22 GCM models in CMIP6 are collected, with the climate variable being rainfall, the period being 1954-2014, and the time scale being daily; the scenario prediction data of 22 GCM models in CMIP6 under the future SSP2-4.5 shared socioeconomic pathway are collected, with the climate variable being rainfall, the period being 2015-2100, and the time scale being daily. The basic information of the 22 GCM models is shown in Table 4.

[0090] Table 4 Overview of selected CMIP6 global climate models

[0091] According to the locations of 20 surface rain gauges, the inverse distance interpolation method is applied to interpolate the gridded rainfall data of GCM historical and future periods to the surface rain gauges.

[0092] (2) GCM simulation capability evaluation and optimization GCM optimization was performed using the GCM simulation capability evaluation method provided in Example 2. The statistical characteristic values ​​and RS score results of the simulated and measured rainfall of 22 GCMs in the Taihu Lake Basin from 1954 to 2014 are shown in Table 5.

[0093] Table 5 Statistical characteristic values ​​and RS scores of each GCM simulated rainfall

[0094] As shown in Table 5, the RS scores of the simulated rainfall in each model are different, with the maximum score being 8.28 and the minimum being 5.38, with a difference of 2.9 between the maximum and minimum. The difference is not large, indicating that the uncertainty of rainfall simulation in different models is small. Among the 22 GCMs, the five models with the best simulation effect for rainfall are EC-Earth3>MRI-ESM2-0>NorESM2-LM>MPI-ESM1-2-LR>CanESM5.

[0095] From the perspective of each statistical characteristic value, there are differences in the characteristic values ​​of each model. Only the simulated rainfall mean of MRI-ESM2-0 is close to the measured value, while the other models are overestimated, and the simulated rainfall of most models is much greater than the measured rainfall. The spatial correlation coefficient between the simulated rainfall and the measured rainfall of each model is good, most of which are greater than 0.8, and the highest value can reach 0.92; the temporal correlation coefficient is not strong, with the highest value of only 0.16, and most models show negative correlation. Only a few of the 22 models can simulate a significant increase in rainfall in the Taihu Lake Basin, and there are large differences between the models.

[0096] (3) Results analysis and comparison The five models with the highest scores in the simulation capability assessment were selected: EC-Earth3, MRI-ESM2-0, NorESM2-LM, MPI-ESM1-2-LR and CanESM5, and their output daily rainfall data were downscaled and analyzed. The daily rainfall data output by each GCM were downscaled using the traditional Delta method, NSRP downscaling model and STNSRP downscaling model, and the daily rainfall data of 20 stations from 1954 to 1996 (rate period) were selected for parameter calibration, and compared with the measured data from 1997 to 2014 (validation period) for verification.

[0097] The GCM simulation capability evaluation method provided in Example 2 was also used to evaluate the simulation capability of the five original models in the validation period and the models downscaled using different methods. The results are shown in Table 6.

[0098] Table 6 Statistical eigenvalues ​​and RS scores of the original model and its downscaling results during the validation period

[0099] As shown in Table 6, the simulation effects of the downscaled models are significantly better than those of the original models, indicating that the three downscaling methods all have the effect of correcting the deviation of the GCM prediction data. Among them, STNSRP has the best simulation effect, followed by NSRP, and the simulation effect of Delta method downscaling is the worst. The three downscaling methods have similar simulation effects on the mean. Delta method only has significant improvements in the mean and standardized root mean square error indicators, and the other indicators are not much different from the original model. This is because the calculation principle of Delta method is to correct only the mean. NSRP and STNSRP have the most obvious improvement on the mean, but there are also greater or less improvements on other indicators, because these two methods use random point processes to simulate precipitation, and correct multiple statistics at the same time, which better simulates the various statistical characteristics of precipitation. Compared with NSRP, STNSRP has a more obvious improvement on the spatial correlation coefficient. This is because STNSRP introduces spatial distribution parameters and makes joint predictions for each station in the basin, which better simulates the spatial correlation of rainfall between different stations and improves the simulation accuracy.

[0100] Taking the EC-Earth3 model as an example, the spatial distribution of the relative errors of the original EC-Earth3 model during the validation period and the EC-Earth3 model downscaled using the Delta method, NSRP, and STNSRP relative to the measured data is shown in the figure. Figure 3 As shown. Figure 3It can be seen that the relative error is significantly reduced after downscaling. The simulation value of the original EC-Earth3 model is systematically larger, with a relative error between 0.27% and 34.35%; the relative error range of the EC-Earth3 model after Delta downscaling is between -6.68% and 12.64%; the relative error range of NSRP downscaling is between -7.07% and 14.00%; the relative error range of STNSRP downscaling is between -6.74% and 13.38%. The correction ability of various downscaling methods for the mean value is comparable, and the spatial distribution of errors after downscaling is relatively similar.

[0101] Taking the EC-Earth3 model as an example, the monthly statistical characteristics of the measured rainfall in the whole basin and Liyang station during the validation period and the simulated rainfall after downscaling using the Delta method, NSRP, and STNSRP are compared. Figure 4 As shown. Figure 4 It can be seen that the four statistics of simulated rainfall after NSRP and STNSRP downscaling are consistent with the measured statistics, and the annual variation process is consistent with the measured value; in contrast, the simulation effect of the Delta method for the mean is also good, but the simulation effect of other statistics is significantly worse than that of NSRP and STNSRP, indicating that the NSRP and STNSRP models can better reflect the annual variation characteristics of rainfall statistics. The simulation results of other stations are similar to those of Liyang Station, and they can simulate the rainfall process well, especially the simulation of heavy rainfall events, showing good simulation capabilities, indicating that the Neyman-Scott method based on random point process can better reflect the annual variation characteristics of rainfall than traditional methods.

[0102] Whether from the perspective of the single station Liyang Station or the whole basin, in terms of monthly statistical characteristics, the simulation effect of the STNSRP model has no obvious difference between the NSRP model and the NSRP model. However, the NSRP model has high requirements for data integrity and cannot be used in areas with insufficient data. The STNSRP model can better reflect the temporal and spatial distribution characteristics of rainfall in the basin through the spatial correlation between the stations, and can also be used to adjust the parameters. The spatial interpolation of the values ​​is used to simulate the rainfall series of the stations with missing data, which has obvious advantages.

[0103] (4) Future rainfall forecast for the Taihu Lake Basin This example uses the results of the five most applicable climate models (EC-Earth3, MRI-ESM2-0, NorESM2-LM, MPI-ESM1-2-LR, CanESM5) in the Taihu Lake Basin after STNSRP perturbation downscaling, as well as the multi-model ensemble results to analyze the possible future climate change in the study area. The prediction time is from 2021 to 2100, and is divided into the short term (2021 to 2050), medium term (2051 to 2075), and long term (2076 to 2100) for discussion.

[0104] 1) Characteristics of future rainfall time changes The linear change trend of annual precipitation in the Taihu Lake Basin from 2021 to 2100 was predicted using the downscaling results of five climate models and multi-model ensemble results. The results are shown in Table 7.

[0105] Table 7 Linear trends of annual precipitation in the Taihu Lake Basin from 2021 to 2100 predicted by various models

[0106] As shown in Table 7, the annual precipitation predicted by each model under the SSP2-4.5 scenario shows an increasing trend except for the MPI-ESM1-2-LR model, but the annual variation is quite different, ranging from -38.1 mm / 100a to 192 mm / 100a, with an average variation of 93.9 mm / 100a. The annual precipitation predicted by the MPI-ESM1-2-LR model shows a slight decreasing trend, with a variation of -38.1 mm / 100a; the CanESM5 and NorESM2-LM models predict a smaller increase, 76.4 mm / 100a and 79.7 mm / 100a respectively; the MRI-ESM2-0 model predicts the largest increase, 192.0 mm / 100a. In general, the annual precipitation in the Taihu Lake Basin will increase in the future, with an increase of about 93.9 mm / 100a.

[0107] The annual distribution of precipitation in the Taihu Lake Basin in the next three periods predicted by the five downscaled models and the multi-model ensemble under the SSP2-4.5 scenario was statistically analyzed and compared with the annual changes of historical measured data from 1954 to 2014. The monthly relative changes are as follows: Figure 5 shown. Figure 5 The difference between the maximum and minimum values ​​indicates the uncertainty of the model prediction, and the median indicates the changing trend of the prediction results. Figure 5It can be seen that the annual distribution of future predicted rainfall is as follows: in the near future, the precipitation in January and September-October will decrease, the precipitation in June-July will decrease slightly, and the remaining months may increase; in the medium future, the precipitation in September-October will decrease, the precipitation in July will decrease slightly, and the remaining months may increase; in the long future, the precipitation in January and November will decrease, the precipitation in August will decrease slightly, and the remaining months may increase. The largest increase in rainfall in the near and medium future is in February, with the median of 14.1% in the near future and 18.8% in the medium term; the largest increase in rainfall in the long future is in February, with a median of 13.8%. The rainfall in November in each period decreases, with the median of -11.0% in the near future, -5.4% in the medium term, and -4.5% in the long term. The uncertainty of rainfall prediction in the flood season (May-September) in each period is relatively large, but the overall increase in rainfall in the flood season is relatively clear, but there are differences in the increase in the predictions of each model. As time goes by, the uncertainty in the overall rainfall forecast increases.

[0108] In general, the annual distribution of rainfall is predicted in the future: precipitation will increase during most of the flood season, and the increase will increase with the passage of time. Although the overall increase is not large, it will also increase the difficulty and intensity of flood control work in the Taihu Lake Basin in the future; the months with reduced precipitation are January and October to November, the dry season in the Taihu Lake Basin; the remaining months may increase.

[0109] The paper focuses on analyzing the precipitation changes in different seasons and years in the next three periods predicted by the five models downscaling and multi-model ensemble, and compares them with the measured data in the historical period (1954-2014). The results are as follows Figure 6 and as shown in Table 8.

[0110] Table 8 Changes in precipitation in different future periods simulated by each model compared with historical periods

[0111] Depend on Figure 6As shown in Table 8, the annual rainfall in each period predicted by each model will mostly show an upward trend, with some slightly decreasing in the near and medium term. With the increase of years, the increase of annual rainfall will also increase. The multi-model ensemble predicts that the annual rainfall in the three future periods will change between 0.5% and 5.4% relative to the measured data of the base period. Summer rainfall will increase in the medium and long term except for a slight decrease of 0.6% in the near term. Spring and winter rainfall will increase in the three periods, with spring rainfall increasing between 3.1% and 9.3%, and winter rainfall increasing between 3.0% and 7.0%. Autumn rainfall will decrease in the near term and increase in the medium and long term, ranging from -3.7% to 5.4%. The EC-Earth3 model predicts that the annual rainfall in the three future periods will decrease in the near term and increase in the medium and long term. Summer rainfall will increase in the medium and long term except for a decrease of 9.1% in the near term. Spring and winter rainfall will increase in the three periods. Autumn rainfall will decrease in the near term and increase in the medium term. The MRI-ESM2-0 model predicts that annual rainfall in the future will decrease slightly in the near and medium term and increase significantly in the medium term; summer rainfall will decrease slightly in the near and medium term and increase significantly in the medium term; spring and winter rainfall will increase in all three periods; autumn rainfall will decrease in both the near and long term and increase in the medium term. The NorESM2-LM model predicts that annual rainfall in the future will increase in all periods, but the increase will not be large; summer and winter rainfall will decrease, and spring and autumn rainfall will increase. The MPI-ESM1-2-L model predicts that annual rainfall in the future will increase in all periods, but the increase will not be large; summer rainfall will decrease slightly in the near term, increase in the medium term, and decrease in the long term; spring and winter rainfall will increase; autumn rainfall will decrease in the near and medium term and increase in the long term. The CanESM5 model predicts that annual rainfall in the future will increase in all periods; summer and winter rainfall will increase; spring and autumn rainfall will decrease in the near term and increase in the medium and long term, but the increase will not be large.

[0112] In general, the seasonal changes in rainfall are predicted in the future: the rainfall in each season will increase overall, with a significant increase in spring and winter, which can alleviate the pressure on water supply to a certain extent; the rainfall in some seasons will show a slight decrease in the near future, mainly concentrated in summer and autumn.

[0113] 2) Spatial distribution characteristics of future rainfall The distribution of changes in the average annual rainfall in the Taihu Lake Basin in the next three periods relative to the baseline period predicted by the downscaling of the five models and the multi-model ensemble under the SSP2-4.5 scenario is as follows: Figure 7 As shown, based on this, the area proportions of different rainfall changes are further statistically analyzed, as shown in Table 9.

[0114] Table 9 The area proportion of rainfall changes in different periods predicted by each model in the future (%)

[0115] Depend on Figure 7It can be seen that the annual rainfall changes of each model in the three future periods all show a trend of gradual increase from east to west, and the increase is mostly greater in the southwest than in the northwest; that is, the rainfall increase in the western mountainous area is greater than that in the eastern plain area, and the mountainous area in western Zhejiang is greater than that in the hilly area in western Lake. Except for the slight decrease in rainfall in the eastern part of the basin predicted by MRI-ESM2-0 and NorESM2-LM in the near future, the rainfall in various regions of the basin predicted by other models in the future will increase. The rainfall predicted by most models will gradually increase with the increase of years, among which the increase of MRI-ESM2-0 is the most significant; but in the far future, the rainfall increase predicted by CanESM5 is not much different from that in the medium term, and may even decrease slightly. Looking at the near future, CanESM5 predicts the largest increase in rainfall, with rainfall changes between 156.9mm and 227.5mm, and rainfall changes between 180mm and 200mm in an area accounting for 60.1% of the basin area. The NorESM2-LM model predicts the most concentrated changes, with rainfall changes between -9.7mm and 0mm in an area accounting for 92.6% of the basin area. In comparison, EC-Earth3 predicts the most dispersed changes, with rainfall changes between 60mm and 80mm in an area accounting for 45% of the basin area. In the medium term, CanESM5 also predicts the largest increase in rainfall, with rainfall changes between 187.0 mm and 236.3 mm, and rainfall changes between 200 mm and 220 mm in the area accounting for 76.8% of the basin area; MPI-ESM1-2-LR predicts the largest change, with rainfall changes between 23.4 mm and 160.2 mm, and rainfall changes between 120 mm and 140 mm in the area accounting for 40.3% of the basin area. In the long term, MRI-ESM2-0 predicts the largest increase in rainfall, with rainfall changes between 168.5 mm and 296.4 mm; NorESM2-LM predicts the most concentrated changes, with rainfall changes between 60 mm and 80 mm in the area accounting for 73.9% of the basin area.

[0116] 3) Characteristics of future extreme precipitation changes This case selected 11 core extreme precipitation indices in three categories proposed by the climate change detection and index expert team to conduct a relatively comprehensive assessment and analysis of the changing characteristics of extreme precipitation in the basin in the future. The conceptual meaning of the extreme precipitation index is shown in Table 10.

[0117] Table 10 Definition of extreme precipitation index

[0118] Using the predicted future rainfall data of the multi-model ensemble MME-5, the annual changes of the extreme rainfall index in the Taihu Lake Basin from 2021 to 2100 relative to the baseline period are shown as follows: Figure 8 The changes of each index relative to the mean value of the base period are shown in Table 11.

[0119] Table 11 Average change of extreme rainfall index in Taihu Lake Basin from 2021 to 2100 relative to the baseline period

[0120] Depend on Figure 8 As shown in Table 11, in the future extreme rainfall in the Taihu Lake Basin, except for the continuous dry days CDD and continuous wet days CWD in the persistence index, which will decrease relatively, the other nine extreme rainfall indices will have an upward trend to varying degrees compared with the baseline period.

[0121] The first type of intensity indexes all show a relative increase. The relative change of Rx1day is greater than that of Rx5day. The increase of SDII is smaller than that of the other two intensity indices, but because SDII represents the change of average rainfall per unit time, there may be extreme rainfall in a certain period of time. Based on the first type of intensity index, the total extreme rainfall in the Taihu Lake Basin will increase in the future because of the concentration of rainfall, and the more extreme rainfall tends to develop in the direction of greater rainfall. The second type of absolute indexes also show a relative increase. The changes of R10, R20 and R30 relative to the baseline period are basically the same, but the relative increase increases successively, that is, the larger the magnitude of rainy days, the more rainy days increase. R95pTOT and R99pTOT have increased significantly in the absolute index, and the relative change is large. From the average change value, the increase of R95pTOT is more significant, reaching more than twice the increase of R99pTOT. Among the third type of persistence index, PRCPTOT showed an upward change, but the relative increase was smaller than R95pTOT and R99pTOT, indicating that the total rainfall was increasing, but the increase was smaller than the extreme rainfall. CDD and CWD showed a downward change compared with the baseline period, which is consistent with the conclusions of the first two types of indexes, indicating that the total precipitation in the Taihu Lake Basin will develop in the direction of increasing rainfall and concentrated rainfall in the future.

[0122] The extreme rainfall index will fluctuate between 2021 and 2100, but it will not affect the overall upward trend compared with the baseline period. In general, rainfall in the Taihu Lake Basin will tend to be concentrated in the future, and the overall trend will be towards increased intensity and frequency of extreme rainfall, and the possibility of severe rainstorms and floods will increase.

[0123] Embodiment 5: This embodiment provides a future rainfall prediction device, including: The data acquisition module is used to obtain the measured rainfall data on the ground, the historical rainfall test data of the optimal GCM, and the future rainfall simulation data; A model building module is used to build a multi-site Neyman-Scott rectangular pulse random rainfall model based on ground rainfall measured data, and optimize the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model to obtain an optimized multi-site Neyman-Scott rectangular pulse random rainfall model; The downscaling module is used to obtain the disturbance change factor based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, and obtain the statistics corresponding to the future rainfall prediction data based on the disturbance change factor; A prediction module is used to input the statistics corresponding to the future rainfall prediction data into the optimized multi-site Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data; Among them, the optimal GCM is the GCM with the highest simulation ability screened out after evaluating the simulation ability of several GCMs.

[0124] The future rainfall prediction device provided in this embodiment can execute the future rainfall prediction method provided in any one of Embodiments 1 to 4, and has the corresponding functional modules and beneficial effects of the execution method.

[0125] Embodiment 6: This embodiment provides a computer system, including: Storage medium: used to store computer programs; Processor: used to execute a computer program to implement the steps of the future rainfall prediction method provided in any one of Examples 1 to 4.

[0126] Embodiment 7: This embodiment provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the steps of the future rainfall prediction method provided in any one of Embodiments 1 to 4 are implemented.

[0127] Embodiment 8: This embodiment provides a computer program product, including a computer program, which, when executed by a processor, implements the steps of the future rainfall prediction method provided in any one of Embodiments 1 to 4.

[0128] Those skilled in the art should understand that the embodiments of the present application can be provided as methods, devices (systems) or computer program products. Therefore, the present application can take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, the present application can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0129] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems) and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram and the combination of the processes and / or blocks in the flowchart and / or block diagram can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0130] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0131] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0132] The embodiments of the present application are described above in conjunction with the accompanying drawings, but the present application is not limited to the above-mentioned specific implementation methods. The above-mentioned specific implementation methods are merely illustrative and not restrictive. Under the guidance of the present application, ordinary technicians in this field can also make many forms without departing from the scope of protection of the purpose of the present application and the claims, which all fall within the scope of protection of the present application.

Claims

1. A method for predicting future rainfall, characterized in that: include: Obtain ground rainfall measured data, historical rainfall test data of the best GCM, and future rainfall simulation data; Based on the measured rainfall data, a multi-site Neyman-Scott rectangular pulse random rainfall model is constructed, and the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model. Based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, the disturbance change factor is obtained, and based on the disturbance change factor, the statistics corresponding to the future rainfall prediction data are obtained; The statistics corresponding to the future rainfall prediction data are input into the optimized multi-station Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data; Among them, the optimal GCM is the GCM with the highest simulation ability screened out after evaluating the simulation ability of several GCMs.

2. The method for predicting future rainfall according to claim 1, characterized in that: The simulation capabilities of several GCMs were evaluated, and the GCMs with the highest simulation capabilities were selected, including: Calculate the simulation capability score of GCM under each evaluation index; Based on the simulation ability scores of GCM under each evaluation index, the total simulation ability score of GCM is calculated; Based on the total simulation ability score of each GCM, the GCM with the highest simulation ability was selected; Among them, the evaluation indicators include relative error evaluation indicators and non-relative error evaluation indicators.

3. The method for predicting future rainfall according to claim 2, characterized in that: Relative error evaluation indicators include: The relative error of the mean of GCM historical rainfall test data to the mean of ground rainfall measured data; The relative error of the coefficient of variation of GCM historical rainfall test data relative to the coefficient of variation of ground rainfall measured data; The normalized root mean square error of the GCM historical rainfall test data relative to the ground rainfall measured data; The relative error between the change trend of GCM historical rainfall test data in the Mann-Kendall trend analysis and the change trend of ground rainfall measured data in the Mann-Kendall trend analysis; The relative error between the trend value of the GCM historical rainfall test data in the Mann-Kendall trend analysis and the trend value of the ground rainfall measured data in the Mann-Kendall trend analysis; Non-relative error evaluation indicators include: The temporal correlation coefficient between the historical rainfall test data of GCM and the measured rainfall data on the ground; Spatial correlation coefficient of GCM historical rainfall test data with ground rainfall measured data.

4. The method for predicting future rainfall according to claim 2, characterized in that: The calculation formula of GCM simulation ability score under each evaluation index is: ； in, Indicates that GCM is The simulation ability score under the evaluation index is Indicates the GCM corresponding The value of the evaluation index, , Indicates the first The maximum and minimum values ​​of the evaluation indicators; The calculation formula for the total simulation ability score of GCM is: ； in, It indicates the total simulation ability score of GCM. The higher the total simulation ability score, the higher the simulation ability. express The corresponding weight, Represents the total number of evaluation indicators.

5. The method for predicting future rainfall according to claim 1, characterized in that: Based on the measured rainfall data, the multi-station Neyman-Scott rectangular pulse random rainfall model is constructed, including: Based on the statistics corresponding to the measured rainfall data on the ground and the fitting weights corresponding to the statistics, the parameters of the multi-station Neyman-Scott rectangular pulse random rainfall model are fitted. Based on the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model, a multi-site Neyman-Scott rectangular pulse random rainfall model is generated; Among them, the statistics include rainfall mean, variance, cross-correlation coefficient of each station, dry day probability and skewness coefficient; Generate a multi-site Neyman-Scott rectangular pulse random rainfall model including: The starting points of rainfall events are randomly generated according to a Poisson process; A random number of rainfall units are generated within each rainfall event. The intervals between the starting points of rainfall units after adjacent rainfall events are independent of each other and follow an exponential distribution. The rainfall units within the same rainfall event follow a Poisson distribution in space, an exponential distribution in radius, and independent duration and rainfall intensity and follow an exponential distribution. The parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model include the characterization parameters of the average interval time between the starting points of adjacent rainfall events, the characterization parameters of the average number of rainfall units within each rainfall event, the characterization parameters of the average interval time between the starting points of rainfall units after adjacent rainfall events, the characterization parameters of the spatial distribution average density of rainfall units within the same rainfall event, the characterization parameters of the average radius of rainfall units within the same rainfall event, the characterization parameters of the average duration of rainfall units within the same rainfall event, the characterization parameters of the average rainfall intensity of rainfall units within the same rainfall event and the scale factor; The value of the scale factor is the quantile of the average rainfall at each station.

6. The method for predicting future rainfall according to claim 1, characterized in that: The parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-site Neyman-Scott rectangular pulse random rainfall model including: Based on the objective function, the parameters of the multi-station Neyman-Scott rectangular pulse random rainfall model are optimized to obtain the optimized multi-station Neyman-Scott rectangular pulse random rainfall model; Among them, the objective function is: ； in, Indicates the first Statistics, express The corresponding annual average value is express The corresponding fitting weight is represents the set of parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model, The multi-station Neyman-Scott rectangular pulse random rainfall model corresponds to The output ground rainfall simulation data corresponds to the Statistics, Indicates the total number of statistics.

7. The method for predicting future rainfall according to claim 1, characterized in that: The calculation formula of the disturbance change factor is: ； in, Indicates The disturbance change factor corresponding to the statistic is , The historical rainfall test data and future rainfall simulation data of the optimal GCM correspond to the Statistics; The calculation formula of the statistics corresponding to the future rainfall forecast data is: ； in, Indicates the future rainfall forecast data corresponding to the Statistics, Indicates the first A statistic.

8. A future rainfall prediction device, characterized in that: include: The data acquisition module is used to obtain the measured rainfall data on the ground, the historical rainfall test data of the optimal GCM, and the future rainfall simulation data; A model building module is used to build a multi-site Neyman-Scott rectangular pulse random rainfall model based on ground rainfall measured data, and optimize the parameters of the multi-site Neyman-Scott rectangular pulse random rainfall model to obtain an optimized multi-site Neyman-Scott rectangular pulse random rainfall model; The downscaling module is used to obtain the disturbance change factor based on the historical rainfall test data and future rainfall simulation data of the optimal GCM, and obtain the statistics corresponding to the future rainfall prediction data based on the disturbance change factor; A prediction module is used to input the statistics corresponding to the future rainfall prediction data into the optimized multi-site Neyman-Scott rectangular pulse random rainfall model to obtain the future rainfall prediction data; Among them, the optimal GCM is the GCM with the highest simulation ability screened out after evaluating the simulation ability of several GCMs.

9. A computer system, characterized in that: include: Storage medium: used to store computer programs; Processor: used to execute the computer program to implement the steps of the future rainfall prediction method described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the future rainfall prediction method described in any one of claims 1 to 7 are implemented.

Images